Reproduce “RD-VIO: Robust Visual-Inertial Odometry for Mobile Augmented Reality in Dynamic Environments” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Jinyu Li, Xiaokun Pan, Gan Huang, Ziyang Zhang, Nan Wang, Hujun Bao, Guofeng Zhang
RD-VIO, a Visual-Inertial Odometry system from Zhejiang University, enhances mobile augmented reality applications by addressing challenges in dynamic environments and during pure rotational motions. The system achieved a 0.136m average RMSE on EuRoC and demonstrated improved robustness over commercial AR platforms in dynamic real-world scenarios.
Key insights
- Leveraging IMU-predicted motion alongside visual consensus and a feature's observation history allows for robust identification and removal of dynamic scene elements without heavy semantic segmentation.
- Explicitly detecting pure rotational motions and managing them with deferred landmark triangulation within a subframe architecture maintains VIO stability and reduces drift even during prolonged low-translation periods.
- Tailoring optimization strategies for different subframe types (normal vs. pure-rotational) and compressing R-frame chains enables real-time, lightweight VIO performance critical for mobile AR devices.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.